This thesis sets out to explore the similarities and differences between post-editing (PE) machine translation (MT) content and traditional translation revision in the professional context of Swiss corporate in-house language services (CILS). Four key aspects have been outlined to guide this comparative analysis, namely linguists’ PE and revision practices, satisfaction, productivity and modifications (edits) made to pre-translated texts during both tasks. A multimethod empirical approach was used, combining data from two different surveys, a field experiment, and a corpus-based analysis of authentic PE and revision assignments.
The data gathered throughout these studies suggest that the boundaries between PE and revision are not as clear-cut as previously perceived. For instance, PE is often performed on fully pre-translated texts, typically as the last step in the workflow. Linguists claim to adopt different approaches depending on whether they are working with human-translated or machine-translated texts. However, in practice, they apply the same reading strategies to these texts. Post-editors tend to repeat the same reading strategies, while revisers adapt them depending on text type, time constraints, and the person who translated the text.
According to data on linguists’ satisfaction with revision and PE assignments, the latter are less fulfilling and slightly less demanding than revision tasks. However, PE – more than revision – enables linguists to create new content and exert control over the text’s final quality. Our data also show that post-editors who received proper PE training report higher satisfaction levels with PE tasks than those who did not receive such training.
The results of the field experiment reveal that linguists are faster during revision than during PE. However, the use of NMT seems to enhance processing speed during PE tasks compared to the previous generation of MT systems. The findings also shed light on the potential quality-related risks of bypassing the verification step performed by a second linguist in PE workflows.
Lastly, the corpus-based analysis shows that CILS linguists edit texts more frequently during PE tasks than revision tasks. Furthermore, human-pre-translated sentences require modifications that are narrower in scope. The task seems to influence the distribution of editing actions, with revision assignments including a higher percentage of deletions compared to PE assignments. Edits in PE assignments focus on addressing mistranslations, while revision assignments mainly involve providing synonyms or rephrasing sentences. The majority of edits performed in both PE and revision tasks are categorised as necessary, although revisers tend to make slightly more optional modifications than post-editors.
In conclusion, this thesis contributes to the growing body of knowledge in translation workplace research, enriching our understanding of contemporary professional translation practices and providing valuable insights for translation pedagogy.
